The availability and exponential growth in online media provides opportunities for understanding and responding to real world challenges. In this paper we investigate the photo quality assessment problem using a large volume of online images retrieved by Google Image Search. To effectively use the big data, we present new approaches that compute discriminative features from a group of relevant images. We also evaluate two popular regression models, support vector regression (SVR) and ranking support vector machine (RankSVM), for their effectiveness in predicting an aesthetic score from the features. Experiments using 99,000 online images provide interesting results. We examine and identify the cases in which online images facilitate the automatic rating task.